Infrastructure planning considerations

Generally, we recommend adding at least 1 of every agent to your instance. On a first-time Agent installation, you can edit the response file and set the DEFAULTAGENTS parameter to 1. This will ensure you have the minimum of required agents installed to your instance on first install.

See the Agents guide for information on how many agents your version of Relativity supports and how to add them to your environment.

Yes, you can scale the number of agents. In addition, you can use more than one agent server in your environment. We suggest allocating one processor core and 1GB of RAM for each additional agent in the environment.

See Scalable Agents and Isolated Scalable Agents for more information on agents that are scalable.

Agent servers must have communication with the SQL Servers (both Primary, Distributed (if applicable), and Invariant SQL) along with the File storage server. See System Requirements for specific hardware requirements on the agent servers.

The web server must communicate with the analytics server via TCP ports 445, 8080, and 8443.

The initial communication from the analytics server to SQL is on ports TCP: 1433 and UDP:1434. By default, communication from the SQL Server to the analytics server occurs on port 8080 or 8443 (REST). However, the response from the analytics server to the SQL Server during a search query can occur on any port, thus all response (ephemeral) ports must remain open. A typical firewall rule would be to allow TCP established from analytics server port 8080 and 8443 to any port range 49152-65535.

The communication between Analytics and SQL is on port 1433.

Using default settings, the average amount of disk space for the Analytics index is equal to 20% of the size of the MDF file of the workspace database. This metric has a high standard deviation and tends to scale up or down linearly. This metric indicates an average amount of disk space usage, but actual indexes may vary considerably in size. The amount of space required depends on the size of the extracted text being indexed, as well as the number of documents, unique terms, and dimensions used to build the index.

The Analytics Index Share houses all of your Analytics data for a particular Analytics server, and it can grow to be very large. We have found that NTFS file systems work for small environments, but if you anticipate running sets of 10 million or more documents through your Analytics Engine, you should use a file system that supports larger files such as exFAT or ReFS. We do not have a recommendation for either file system, so you must determine which is the better fit for you.

No. An analytics server may only reference one disk location for the server’s analytics indexes and structured analytics sets.

Yes. Analytics servers are scalable and are available as Servers in Relativity. You can allocate the Analytics servers to different workspaces via Resource Pools. Additionally, you can dedicate Analytics servers to Analytics Indexing, Structured Analytics, or both feature sets. For more information on configuring these options, see Adding an Analytics server.

Relativity is a web-based review platform that rests on a distributed, grid-based computational model.

You can use an Ethernet connection for the servers in your Relativity installation. Minimal connectivity speeds between SQL and the storage location are 4Gbps. See System Requirements for specific network connectivity requirements.

We advise against this server configuration. When users are heavily utilizing the agents, this configuration causes performance problems. It also makes troubleshooting performance issues difficult.

Yes, Relativity can run over a WAN if it's set up for pass-through authentication. All authentications must pass through to Relativity.

For a default named SQL Server, switching port 1433 won't work.

For a named instance of SQL Server, you can use a static port for SQL communication. By opening up port 1434 and setting the SQL Server to use another port (for example, 1435), SQL will begin to communicate using port 1435.

However, for Relativity to work and communicate effectively over the chosen static port, UDP port 1434 also needs to remain open. Relativity uses this port for the initial connection to the SQL Server and then begins to use port 1435. Once SQL has been configured to use a static port such at 1435, if port 1434 isn't open, certain actions in Relativity might break. Some of the places where it breaks include upgrading and the Relativity viewer.

Indexes for the three searching methods offered in Relativity are stored in the following locations:

• Keyword Search - Relativity stores indexes for these searches in the SQL database.
• dtSearch - Relativity stores dtSearch indexes in the file repository.
• Analytics - You should store indexes and structured analytics sets on a drive letter attached to the Analytics server. This should have similar storage speed and connectivity of SQL Server.

Yes, you can configure your Relativity environment with multiple repositories and dtSearch indexes as necessary.

No, a Windows installation isn't required, but you must provide Relativity with the path to the location where the files reside.

We don't recommend using thin virtual provisioning in VMware for the following reasons:

• Host system on SAN - SAN vendors differ in their support of thin virtual provisioning. If a device throttles connectivity speeds when certain thresholds are met, monitor these thresholds and configure alerts to notify system admins of these events.

  Provisioning can cause Relativity to reach capacity and a disk grow (zeroing) event to occur. If provisioning uses zeroing to grow the disks, performance is throttled at as little as 50 percent of expected disk IO.

• Exhaustion of shared storage pools - Relativity can exhaust storage pools when LUNs are over-provisioned and data writes to them. To avoid this situation, you must put procedures, warnings, and protocols in place. Use thick provisioning and zero disks to prevent a rogue host from writing large amounts of data that infringes on other storage users.
• Mismatched block sizes - each implementation of thin provisioning uses a different block size. For example, HDS uses 42 MB on the USP, EMC uses 768 KB on the DMX, IBM uses variable sizes from 32 KB to 256 KB on the SVC, and 3Par uses blocks of 16 KB. While the reasons for these practices vary, file systems created on thin provisioned LUNs don't typically have a matching block size structure. This makes thin provisioning a bad choice for SQL data.
• Mismatched Block Sizes Effects SQL Efficiency - by default, Windows NTFS uses a maximum block size of 4 KB for large disks—unless it has been configured for a specific size. The mismatch between the block sizes for thin provisioning and the file system causes SQL performance problems on these systems when data is created, amended, and deleted over time. Consequently, thin provisioning negatively affects the efficiency of the SQL database when performing these operations.

We do not have any specific storage speed or IOPS requirements for Relativity. However, you can use the following guidelines when selecting storage devices:

• Fast storage - select the fastest available storage for SQL data.
• Minimal latency - per Microsoft best practices for database SAN LUNs, SQL Server disk read/write latency should generally be 20 ms or less. The SQL Server disk read/write latency for transaction log SAN LUNs should 5 ms or less following Microsoft best practices.
• Optimal performance configuration - follow the guidelines from your storage device provider to ensure that it's configured for optimal performance. In addition, work with your device provider or follow Microsoft best practices to determine where different SQL data should live. Ensure that the link between the SQL Server and storage device is configured correctly for optimal performance by consulting your device provider. This setup becomes more complicated when virtualizing SQL.
• Consider iSCSI - while fiber has traditionally been considered a better option for enterprise class storage, iSCSI protocols have changed that to some extent. Even though most fiber cards now provide 4 GB of connectivity while Ethernet is limited to 1 GB, the speed differences aren't very pronounced. Fiber tends to cost more, but you want a dedicated Ethernet switch for iSCSI traffic. We have multiple clients using iSCSI without any problems.
• Storage maintenance - when you purchase and evaluate storage, be sure to discuss a disaster recovery plan for it, the qualifications of the staff required to manage it, and how it will support continued growth of the environment.
• Space requirements - on average, every 250,000 records in a Relativity workspace consumes about 20 GB of MDF disk space. As a rough metric, the total SQL data for a workspace is about one-third the total file size (natives and images) of that workspace.
• Auto-tiering - many new SAN devices can automatically shift data (auto-tier) across different types of available storage based on the activity levels of the stored data. You can use these devices to maximize performance with minimal manual intervention, reducing IT-related costs and maximizing use of available storage.

  We use Compellent storage, while many of our partners have recently purchased EMC VNX devices. Other devices, such as NetAPP and IBM, also use some type of auto-tiering.

• Windows Server 2016
• Windows Server 2019
• Windows Server 2022

The Virtual Machine Communication Interface (VMCI) is an infrastructure that provides fast and efficient communication between a virtual machine and the host operating system. It also provides fast and efficient communication between two or more virtual machines on the same host.

You can enable your application to use VMCI programmatically. VMCI provides greater inter-guest communication when virtual machine guests share a host. We don't offer built-in support for this feature. However, you may want to research existing tools for porting specific TCP endpoints to VMCI sockets. See the VMCI Sockets Documentation.

By default, imaging jobs take higher priority over processing jobs. However, you can access the management to change the priority of running jobs. There is never a need to manually configure things. If a processing worker has nothing to do, it asks the Queue Manager to give it something to do. Any processing worker can work on any job, regardless of job type. It's uncommon for even a single worker to have multiple threads working on both imaging and processing simultaneously.

Yes, you should back up processing databases.

We recommend a decent amount of HD space on the worker machine for processing, because up to 16 threads can perform processing on each worker. Some of the files each worker processes could be quite large. 250 GB is a reasonable amount of space to perform temporary work before files get transferred to the network. The space is strictly used for temporary storage of data while workers are processing.

Use Relativity Processing to ingest raw data directly into your workspace for eventual search and review without the need for an external tool. Refer to the Environment Optimization Guide for a more detailed explanation on the stages of Processing.

Yes, Relativity supports virtualized processing SQL Servers as long as the server meets the minimum requirements.

Yes, you can specify the type of jobs a worker can perform by editing the Worker designated work field on the worker server. You can configure a worker for one or more or all of the following jobs:

• Processing
• Save as PDF
• Native imaging

See the Legal Hold compatibility matrix for version support details.

You can only upgrade from version 2.1.127.5 and above to 3.0. Any version beneath that, including 1.0, does not have a direct upgrade path. If a client really needs to maintain their data during an upgrade from an older version, reach out to PM.

You can see it all in one place from the Configuration tab or look at the installed Applications.

You can install to any workspace that doesn"t have Processing already installed.

We recommend starting out with a single agent because it doesn’t use a lot of system resources. You can add more agents, if necessary.

There is no authentication done when accessing the Portal, but the Portal link is unique to every custodian. To prevent unauthorized use (via email forwarding of the link), you can specify the maximum number of times the link can be used or specify a link expiration date.

No, it doesn't.

Yes, you can configure your Relativity environment with multiple repositories and dtSearch indexes as necessary.

Each Relativity dtSearch search agent should have its own server or virtual machine, similar to existing dtSearch WebAPI web servers. System admins have the ability to add the search agent to servers with other Relativity agents. However, this is not advisable as one dtSearch search agent will be able to utilize all available processor cores on a server. Therefore, each Relativity agent server that is designated to be a dtSearch search agent server should only have one dtSearch search agent and nothing else.

The dtSearch indexes grow to approximately 25% of the total database size. For example, if you had 10 TB of total files, the database would be 3-4 TB, and you would need about 1 TB for the index storage.

The default size of the sub-index is controlled through an instance setting. It can also be adjusted per each index creation through a setting on the index creation configuration page. The default value will is 250,000 documents. Larger data sets may benefit from having the sub-index size increased to 500,000 documents.

Yes, you can use a solid-state drive (SSD) on the SQL Server to improve performance.

We don't recommend storing dtSearch indexes on the SQL Server. Other than this location, you can store the indexes in the file repository location.

Multiple files provide better I/O performance and less contention on the global allocation structures. For higher end systems with more than 16 cores, Microsoft suggests beginning with eight database files and adding more if there is still contention.

There are scripts available to measure contention. Most Relativity environments work well with eight data files. The Environment optimization guidedescribes several issues that you may want to consider when configuring the tempdb.

Yes, the focus on connection speed should be between the SQL and the storage system. A SQL Server on 1 GB iSCSI isn’t sufficient. It constrains you during backups, index rebuilds, table scans, etc. These large sequential operations can easily saturate a 1 GB pipe, and storage quickly becomes a bottleneck. The minimum requirement we have is 4GBPS, but if possible, we recommend moving this up to 8-10GBPS.

With the use of multiple active SQL Servers, you can distribute your workspace databases across more than one SQL instance.

Yes, another instance can be added to this configuration and then moved to a second node or converted to an active/active cluster.

If there is a need to change the name of the primary Relativity SQL instance during the reconfiguration, Customer Support will assist you with updating references to it in Relativity. (The primary Relativity SQL instance houses the EDDS database.)

For information about SQL partition sizes, see System Requirements. That document includes sample partition sizes for the databases, logs, tempdb, and other features.

No, you should never activate the Priority Boost setting.

SSL transit is the most commonly used database encryption method in practice. Some clients have been known to use Transparent Data Encryption, which is available in the enterprise edition of SQL Server.

The physical memory limitations are shown below:

• Windows Server 2016 Datacenter & Standard is 24TB
• Windows Server 2019 Datacenter & Standard is 24TB
• Windows Server 2022 Datacenter & Standard is 48TB

The maximum supported RAM in GB for SQL Server standard addition for SQL Server 2017, 2019, & 2022 is 128GB. For more information, see System Requirements.

The database is small at the outset. It grows as it ages (from audits) and as data is loaded into it. The amount of SQL data created during a load can be anywhere from 1-2 times the size of the load file being loaded. This can possibly be even more if a lot of ANSI data is loaded into Unicode fields.

We require both a Windows Relativity service account and a SQL Server account (EDDSDBO), but the EDDSDBO server account is the only one that connects to the database itself.

The Windows Relativity service account requires permissions to the shared BCP Path folder on the SQL Server so that it can drop data there for loading.

What are the recommendations for virtualizing servers in a Relativity installation?

We recommend virtualizing all Relativity roles with the exception of SQL. Any hypervisor is supported.

Note: Virtualizing SQL may simplify meeting High Availability (HA) and Disaster Recovery (DR) requirements. However, it can introduce additional layers of complexity when properly configuring and troubleshooting performance and stability related issues.

We do not have any special recommendations for using VMware with Relativity.

Use the following guidelines to determine the number of web servers required for your Relativity installation:

• One web server dedicated for importing and exporting with Relativity Desktop Client (RDC).

• The Web (Application) role supports 50 simultaneously logged in users for every 16 processor cores and 32 GB RAM.

To configure the servers for DMZ access, you must open the following ports between the web server in the DMZ and all other servers in your environment:

• 445 (Microsoft-DS Active Directory, Windows shares)
• 1433 (MSSQL Server TCP)
• 1434 (MSSQL Monitor UDP)
• 8080 (Analytics)
• 8443 (Analytics)
• 443 (SSL) open to the outside
• 25 (SMTP)
• TCP 9090 (Secret Store)
• TCP 10000 (Secret Store)

Only the web servers should be in the DMZ.

• Windows load balancing(single affinity) - after a user’s session begins on a single web server, it must remain on that server, or errors occur. In Relativity, a session state is stored in proc.
• RULB (Relativity.User.Load.Balancing) - RULB requires each included web server to have its own externally facing URL. To enable RULB you would go into the Servers tab of the admin page and edit a web server you want to include in the RULB. You then enter the URL for that web server you are editing.

  Note: Once you click Save, Relativity checks the URL to see if it accessible so make sure the path is set up beforehand.

  RULB works well with NLB and needs no extra configuration. All you have to do is turn on RULB, and Relativity does the rest. You probably won't see the NLB URL in IE change, but users should land on different web servers.

  RULB also works with SSL certificates.

• Round robin - this is the default load balancing method. Round robin mode passes each new connection request to the next server in line, eventually distributing connections evenly across the array of machines being load balanced. Round robin mode works well in most configurations, especially if the equipment that you're load balancing is roughly equal in processing speed and memory.
• Least connections (member) and least connections (node) - the least connections methods are relatively simple. The LTM system passes a new connection to the node that has the fewest current connections. The least connections methods work best in environments where the servers or other equipment being load balanced have similar capabilities.

  These are dynamic load balancing methods. They distribute connections based on aspects of real-time server performance analysis, such as the current number of connections per node or the fastest node response time. Load balancing calculations may be localized to each pool (member-based calculation) or they may apply to all pools of which a server is a member (node-based calculation).

• Priority-based member activation - you can load balance traffic across all members of a pool or across only members that are currently activated according to priority number. In priority-based member activation, each member in a pool is assigned a priority number that places it in a priority group designated by that number. 

  With all pool members available (meaning they are enabled, marked up, and have not exceeded their connection limit), the LTM system distributes connections to all members in the highest priority group only. That is, the group designated by the highest priority number. The Priority Group Activation value determines the minimum number of members that must remain available for traffic to be confined to that group. If the number of available members in the highest priority group drops below the minimum number, the LTM system also distributes traffic to the next highest priority group, and so on.

RULB (Relativity.User.Load.Balancing) or NLB are the most commonly used load balancers.

What load balancing algorithms are available as Predictor options?

The following load balancing algorithms are available as Predictor options:

• Round robin - distributes connections evenly among the servers (default setting).
• Weighted round robin - enables priorities to be assigned to certain servers. The higher a server's weight, the more connections the load balancer sends to it.
• Least Connections - directs a service request to the server with the fewest open connections.

The default ports utilized for secret store are TCP 9090 (the service port) and TCP 10000 (the nonce port). All Relativity Servers must be able to communicate with Secret Store over these ports and these ports are configurable.

No, it is not possible to recover secrets from the store without the unseal key. In this scenario, it is recommended that the existing secret store database be backed up, deleted and that Secret Store be reinstalled to create a new database and generate a new unseal key. Secrets should be reimported into the store at this point.